bug#7848: 23.2.91; Can't build with MinGW

bug#7848: 23.2.91; Can't build with MinGW

[Kazuhiro Ito]

When I build emacs-23 branch at present with MinGW, below error
occurs while linking.

> oo-spd/i386/temacs1.a(image.o):image.c:(.text+0xa8fa): undefined reference to `SET_SYMBOL_VAL'
> collect2: ld returned 1 exit status
> make[1]: *** [oo-spd/i386/temacs.exe] Error 1
> make[1]: Leaving directory `/f/build/emacs-23/src'
> make: *** [all-other-dirs-gmake] Error 2

SET_SYMBOL_VAL is defined in src/lisp.h on trunk but not on emacs-23
branch.

-- 
Kazuhiro Ito

bug#7848: 23.2.91; Can't build with MinGW

[Eli Zaretskii]

> Date: Sat, 15 Jan 2011 16:34:08 +0900
> From: Kazuhiro Ito <kzhr@d1.dion.ne.jp>
> Cc: 
> 
> When I build emacs-23 branch at present with MinGW, below error
> occurs while linking.
> 
> > oo-spd/i386/temacs1.a(image.o):image.c:(.text+0xa8fa): undefined reference to `SET_SYMBOL_VAL'
> > collect2: ld returned 1 exit status
> > make[1]: *** [oo-spd/i386/temacs.exe] Error 1
> > make[1]: Leaving directory `/f/build/emacs-23/src'
> > make: *** [all-other-dirs-gmake] Error 2
> 
> SET_SYMBOL_VAL is defined in src/lisp.h on trunk but not on emacs-23
> branch.

Sorry.  Please try replacing this:

  SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (PNG_LIBPNG_VER));

with this:

  XSYMBOL (Qlibpng_version)->value = make_number (PNG_LIBPNG_VER);

and make a similar change in this line:

  SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (-1));

I cannot test this where I'm typing this mail, but if you tell me that
these two changes fix the problem, I will commit this change to the
release branch.

bug#7848: 23.2.91; Can't build with MinGW

[Stefan Monnier]

> Sorry.  Please try replacing this:

>   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (PNG_LIBPNG_VER));
[...]
>   XSYMBOL (Qlibpng_version)->value = make_number (PNG_LIBPNG_VER);

I would recommend to stay away from such low-level code unless you're
writing code that tries to implement part of the binding-semantics
of Emacs.  I.e. just use Fset.


        Stefan

bug#7848: 23.2.91; Can't build with MinGW

[Kazuhiro Ito]

> > When I build emacs-23 branch at present with MinGW, below error
> > occurs while linking.
> > 
> > > oo-spd/i386/temacs1.a(image.o):image.c:(.text+0xa8fa): undefined reference to `SET_SYMBOL_VAL'
(snip)
> > SET_SYMBOL_VAL is defined in src/lisp.h on trunk but not on emacs-23
> > branch.
> 
> Sorry.  Please try replacing this:
> 
>   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (PNG_LIBPNG_VER));
> 
> with this:
> 
>   XSYMBOL (Qlibpng_version)->value = make_number (PNG_LIBPNG_VER);
> 
> and make a similar change in this line:
> 
>   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (-1));
> 
> I cannot test this where I'm typing this mail, but if you tell me that
> these two changes fix the problem, I will commit this change to the
> release branch.

Thanks, I confirmed above changes fixed my problem (not tested in the
case of no PNG support).

-- 
Kazuhiro Ito

bug#7848: 23.2.91; Can't build with MinGW

[Eli Zaretskii]

> Date: Sun, 16 Jan 2011 15:16:31 +0900
> From: Kazuhiro Ito <kzhr@d1.dion.ne.jp>
> Cc: 7848@debbugs.gnu.org
> 
> > Sorry.  Please try replacing this:
> > 
> >   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (PNG_LIBPNG_VER));
> > 
> > with this:
> > 
> >   XSYMBOL (Qlibpng_version)->value = make_number (PNG_LIBPNG_VER);
> > 
> > and make a similar change in this line:
> > 
> >   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (-1));
> > 
> > I cannot test this where I'm typing this mail, but if you tell me that
> > these two changes fix the problem, I will commit this change to the
> > release branch.
> 
> Thanks, I confirmed above changes fixed my problem (not tested in the
> case of no PNG support).

Thanks, I installed these changes on the emacs-23 branch.

bug#7848: 23.2.91; Can't build with MinGW

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: Kazuhiro Ito <kzhr@d1.dion.ne.jp>,  7848@debbugs.gnu.org
> Date: Sat, 15 Jan 2011 23:41:51 -0500
> 
> > Sorry.  Please try replacing this:
> 
> >   SET_SYMBOL_VAL (XSYMBOL (Qlibpng_version), make_number (PNG_LIBPNG_VER));
> [...]
> >   XSYMBOL (Qlibpng_version)->value = make_number (PNG_LIBPNG_VER);
> 
> I would recommend to stay away from such low-level code unless you're
> writing code that tries to implement part of the binding-semantics
> of Emacs.  I.e. just use Fset.

Using Fset here would be a terrible overkill, IMO.  I'm not even sure
I understand all of its semantics, and not sure it will work correctly
in the context of an init_* function that is called when Emacs is
dumped.  OTOH, direct assignments of symbol's value like above _are_
used in several init_* functions and elsewhere.

Also, I wanted to un-break the release branch as fast as I could.

Having said all that, feel free to change what I did to use Fset, both
on the branch and on the trunk.

bug#7848: 23.2.91; Can't build with MinGW

[Stefan Monnier]

>> I would recommend to stay away from such low-level code unless you're
>> writing code that tries to implement part of the binding-semantics
>> of Emacs.  I.e. just use Fset.
> Using Fset here would be a terrible overkill, IMO.

Overkill?  For code that's not even inside a loop?  What are you talking about?

> I'm not even sure I understand all of its semantics,

The semantics of Fset are the same as the ones of `setq' which you use
days-in days-out in Elisp.  I.e. it's the semantics you know best,
whereas the semantics of SET_SYMBOL_VAL are a lot more tricky.

> and not sure it will work correctly in the context of an init_*
> function that is called when Emacs is dumped.

Yes, it'll work just fine, pretty much in any context (the only tricky
context in this respect is when/before the `nil' symbol&variable and the
`unbound' value are built, AFAIK).

> OTOH, direct assignments of symbol's value like above _are_
> used in several init_* functions and elsewhere.

I've fixed some of those in the past, and feel free to fix more of them.

> Also, I wanted to un-break the release branch as fast as I could.

I'm not complaining, at all.  Just giving advice.


        Stefan

bug#7848: 23.2.91; Can't build with MinGW

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: kzhr@d1.dion.ne.jp,  7848@debbugs.gnu.org
> Date: Sun, 16 Jan 2011 10:21:53 -0500
> 
> >> I would recommend to stay away from such low-level code unless you're
> >> writing code that tries to implement part of the binding-semantics
> >> of Emacs.  I.e. just use Fset.
> > Using Fset here would be a terrible overkill, IMO.
> 
> Overkill?  For code that's not even inside a loop?  What are you talking about?

Look at its body, man.  All I want is to assign a simple value to a
simple global-scope variable.  Look how many unrelated use-cases the
implementation of Fset handles.  How can a mere mortal who was never
privy to the intimate details of whatever-local variables possibly
grasp what is going on there?  How can I be expected to be able to
install a quick and safe fix on a release branch with all that
complexity that is utterly unrelated to what I want to do?

If you want to avoid code that knows too much about variable
internals, we should have a much simpler utility function to do this
kind of job in such special cases.  SET_SYMBOL_VAL looked like such a
utility.

> > I'm not even sure I understand all of its semantics,
> 
> The semantics of Fset are the same as the ones of `setq' which you use
> days-in days-out in Elisp.

I use no Lisp at all during dumping.  There are certain things one
shouldn't do at that time.  My everyday experience in Emacs does not
help me in this case.

> I.e. it's the semantics you know best, whereas the semantics of
> SET_SYMBOL_VAL are a lot more tricky.

If it is tricky, please consider putting a suitable commentary near it
to explain the trickery, and also tell why this macro should be used
with care, or be avoided, or whatever.

> > and not sure it will work correctly in the context of an init_*
> > function that is called when Emacs is dumped.
> 
> Yes, it'll work just fine, pretty much in any context (the only tricky
> context in this respect is when/before the `nil' symbol&variable and the
> `unbound' value are built, AFAIK).

And I'm supposed to learn this from where, exactly? And how do I know
whether any specific code runs after that critical point?

That's precisely why I avoided Fset: because I could not convince
myself that it is always safe to use it in a function that runs during
dumping.  If Fset's implementation was simpler, I could have
understood that myself.

> I'm not complaining, at all.  Just giving advice.

Thanks for the advice.